BIOS 740: STATISTICAL LEARNING AND HIGH-DIMENSIONAL DATASpring, 2011• COURSE DESCRIPTION (3 credit hours)The course gives an introductory overview of statistical learning methods with or without high-dimensional data. It consists of three major components: learning methods, learning theoryand methods for high-dimensional data. The first part on learning methods provides a completereview of supervised learning methods (discriminant analysis, kernel methods, nearest neigh-borhood, tree methods, neural network, support vector machine, random forest, and boostingmethods) and unsupervised learning methods (principal component analysis, factor analysis,cluster analysis, multidimensional scaling, self organizing map). R-functions and real data demoare used for illustration. The second part on learning theory provides some foundational theoryfor statistical learning methods and it covers theories of Bayesian error, concentration inequal-ities, VC-theory, risk bound and etc. The last part of the course focus on current statisticalmethods with high-dimensional data including dimension reduction, variable selection and mul-tiple testing, along with a number of real applications.• MEETING TIME 11:00-12:15, Monday and Wednesday @ Room MCG2305• CLASS WEBSITE http:\\ www.bios.unc.edu\ ∼dzeng\Bios740.html• LECTURE NOTES AND TEXTBOOKS– Lecture notes (can be downloaded from the class website)– The Elements of Statistical Learning: Data, Mining, Inference, and Prediction, by Hastieet al. (downloadable from their website)– A Probabilistic Theory of Pattern Recognition, by Devroye et al.– Learning with Kernels: Support Vector Machines, Regularization, Optimization and Be-yond, by Schlkopf and Smola– Principles and Theory for Data Mining and Machine Learning, by Clarke et al.• INSTRUCTORDr. Donglin ZengOffice: 3103B McGavran-Greenberg BuildingEmail: [email protected]: (919)966-7273Office hours: 1-3 Friday• GRADING SYSTEMThere will be no exams for this course. Homework assignments will be given occasionally. Afinal project, which can be research-oriented work, or pap er review, or real data application,will be required. Final grades will be based on the performance of homework and the finalproject including the final presentation. For the final project, students are encouraged to meetthe instructor to discuss choices around the middle of the semester.• TOPICS TO BE COVERED1. Supervised Learning (10 lectures)– Statistical decision theory– Direct learning: parametric methods1– Direct learning: Semi-Nonparametric methods– Direct learning: Nonparametric methods– Indirect learning2. Unsupervised Learning (2 lectures)– Principal component analysis– Latent component analysis– Multidimensional scaling– Cluster analysis3. Learning Theory (10 lectures)– Bayes error– Consistency of direct learning methods– Consistency of indirect learning methods– Convergence rates– Classification error estimation– Concentration inequalities4. High-dimensional Data (6 lectures)– Dimension reduction– Variable selection– Multiple testing– Application-specific methods• OTHER INFORMATION– Teaching tool will be mainly based on the use of the projector, sometimes with the help ofchalkboards or handouts.– I will be out of town March 21-30 so we need to make up 4 classes in April. My initial planis to have them on Fridays of April but this is subject to change.– The classes on April 20, 25, 27 and May 2 are reserved for students to present their